Pre-PAW
2026 WCPO Bigeye Assessment
Identify issues and priorities
Kyuhan Kim | SPC-OFP
13 November 2025
1. Review 2023 Issues - What went well, what didn’t
2. Identify Priorities for 2026 - Must address vs nice-to-have
3. Key Decisions - Spatial: 6R vs 9R? - Tag mixing: 1Q/2Q/3Q? - Indonesia catch treatment
WCPO BET Annual Catches (1952-2021)
By Gear Type:
Trend: Relatively stable since 2000
Tropical regions dominate:
Catch-conditioned model (Day et al. 2023): - Reduced from 11,421 to 3,067 parameters - More stable convergence
sdmTMB CPUE standardization (Teears et al. 2023): - Spatial-temporal modeling framework - Replaced VAST approach
Lorenzen natural mortality (Lorenzen 1996): - Size/age-dependent M scaling - More realistic mortality patterns
Internal growth estimation (Farley et al. 2020): - Von Bertalanffy parameters estimated - 1,004 otoliths incorporated
Punt et al. (2023) (Punt, Maunder, and Ianelli 2023): - Reviewed 2020 YFT assessment - SC19 decision: Apply to BET where practical
Key recommendations implemented: - Revised tagger effects modeling (Peatman, Scutt Phillips, J., and Nicol 2023) - Tag mixing period as grid axis - Size composition reweighting methods - Growth-M correlation handling
Extensive jittering (Day et al. 2023): - Up to 60 runs per model to find global optimum - Diagnostic model achieved positive definite Hessian - 48 of 54 grid models converged
SB/SBF=0 = 34% (Day et al. 2023): - Just below LRP (40%) - Median across 54-model grid
Fishing mortality above FMSY: - Overfishing occurring - Primarily from purse seine associated sets
Recruitment stable since 1960: - ~120 million fish/year - No evidence of recruitment impairment
Time series: 1952-2021 (70 years)
Size compositions: 41 fisheries
Tagging: ~67K BET (RTTP, PTTP, JPTP)
CPUE: Longline indices (9 regions)
54 models exploring (Day et al. 2023): - Tag mixing: 1Q, 2Q, 3Q (most influential) - Steepness: 0.65, 0.80, 0.95 - Size comp weighting: 3 methods - Age data weighting: 6 levels
High parameter uncertainty: - Correlations between M, growth, recruitment - Sensitivity to starting values
Key data limitations: 1. Limited age data (1,004 otoliths) 2. Spatial structure complexity (9 regions) 3. Size composition conflicts with tagging 4. Tag mixing period uncertain
SC19 Evaluation (WCPFC 2023):
“More rigorous… but model instability remains”
Result: 14 research recommendations
Continue age data collection; evaluate growth/mortality variability; explore regional traits
Biological data (Hamer et al. 2023): - Low tropical <-> temperate movement - High mixing within tropical/temperate zones - Size composition clustering supports simpler structure
Model performance: - 6-region used 2005-2011 successfully - Faster runtime, fewer parameters - Better convergence stability
Simulation study: 6-region vs 9-region performance; “areas-as-fleets” approach
2024 catch spike: Large increase in handline catch — major uncertainty in historical allocation
Comprehensive review of size composition representativeness and Region 7 fisheries
RTTP (1989-92): 8.6K; PTTP (2006-20): 59K; JPTP (2000-20): added in 2020
Develop diagnostics; test 1Q/2Q/3Q; quantify uncertainty (Peatman, Scutt Phillips, J., and Nicol 2023)
No iterative reweighting in 2023: - Fixed weights (no iterative weighting due to time constraints) - Data conflicts unresolved
Review representativeness; investigate inversion
Tropical regions (3,4,7,8): unfished SSB increases over time — model artifact?
Investigate model specifications causing this trend
Sensitivity to initial conditions — Models achieve suboptimal local minima
Diagnostic model: - Primary jittering (60x): Improvements of tens to hundreds of likelihood units - Stock status shifts: 5%+ in SB/SBF=0
Grid models: - Before: 46/54 with PDH → After: 48/54 with PDH - Every model improved through jittering
SC19-#9, #13: Identify correlated/sensitive parameters; strategies to decrease starting value sensitivity
“Model convergence is a significant issue for these complex models… jittering is an important process to refine and improve the best solution.”
“Jittered solutions achieved improved objective function values and notable changes to derived management quantities. This was cause for considerable concern.”
“Every grid model improved through jittering — none was at a global minimum, even those 46 with positive definite Hessians.”
Model structure: - Catch-conditioned age-structured model - 9-region spatial framework - Quarterly time steps (1952-2021)
Key methodologies: - sdmTMB CPUE standardization - Lorenzen M-at-age scaling - Internally estimated von Bertalanffy growth - Extensive likelihood profiling
YFT review recommendations applied (Punt, Maunder, and Ianelli 2023): - Catch-conditioned framework - Internal growth estimation - Enhanced CPUE standardization - Improved convergence diagnostics
“The 2023 assessment applied a more rigorous approach to obtain a stable converged model than that applied in the 2020 assessment; however, model instability appears to remain.”
— SC19 Outcomes (WCPFC 2023)
Strengths recognized: - Comprehensive uncertainty grid - Extensive sensitivity testing - Improved convergence diagnostics
Remaining concerns: - Parameter convergence issues - Structural uncertainties
Address SC recommendations while: - Maintaining methodological strengths - Improving model stability - Expanding uncertainty characterization
Operational catches (2022-2024): - Longline fleets - Purse seine (associated/unassociated) - Domestic fisheries (Indonesia, Philippines, Vietnam)
Indonesia handline issue: - Large catch increase in 2024 - Gear allocation uncertainty (small -> large handline) - Historical series implications
CPUE indices (SC19-#2, #3) (WCPFC 2023): - sdmTMB standardization updated - Effort creep investigation - Additional index series tested
Reweighting methods (SC19-#11, #12) (WCPFC 2023; Punt, Maunder, and Ianelli 2023): - Francis method - Revisit Dirichlet-multinomial - Flag/fishery-specific weighting - Weight vs tag conflict investigation
New tag returns (SC19-#1) (WCPFC 2023): - PTTP 2022-2024 recaptures - Tag-reporting rate updates (Peatman, Scutt Phillips, J., and Nicol 2023) - Mixing assumption refinement
Growth estimation (SC19-#7, #10) (WCPFC 2023): - Continue internal estimation - 1,004 otolith CAAL dataset - Regional/seasonal variation testing - Growth-mortality correlation
SC19-#5: Unfished SSB trend diagnostics
SC19-#6: Yield-per-recruit by fishery sector
1. Steepness (Day et al. 2023): - 0.65 (pessimistic) - 0.80 (base) - 0.95 (optimistic)
2. Tag mixing (WCPFC 2023; Peatman, Scutt Phillips, J., and Nicol 2023): - 1 quarter - 2 quarters (base)
3. Size composition weighting (Peatman et al. 2023): - 10 (high weight) - 20 (base) - 40 (low weight)
4. Age composition weighting (Day et al. 2023): - 0.50 (low weight) - 0.75 (base) - 1.00 (high weight)
Spatial structure (SC19-#4) (WCPFC 2023; Hamer et al. 2023): - 6-region vs 9-region - Simulation study results (cross-fit tests: simulate from 9 -> fit 6, and vice versa)
Tag mixing period (SC19-#14): - 3 quarters (extended) - Piner plot validation
Structural: - Natural mortality scaling - Selectivity patterns
Data treatment: - Indonesia catch scenarios - CPUE effort creep - Size comp weighting schemes
SC19-#9 (WCPFC 2023): Identify problem parameters - Profile correlated parameters - Test parameter fixing options - Reduce dimensionality where possible
SC19-#13 (WCPFC 2023): Improve stability - Multiple starting values (jittering) (Day et al. 2023) - Phased parameter estimation - Hessian diagnostics at each step
Consider reducing complexity (Punt, Maunder, and Ianelli 2023): - Spatial structure (6R vs 9R) (Hamer et al. 2023) - Fishery aggregations - Selectivity parameterization
Retrospective analysis (Punt, Maunder, and Ianelli 2023): - Peel test (5-10 years) - Mohns rho calculation - Pattern identification
Tag diagnostics (SC19-#14) (WCPFC 2023; Peatman, Scutt Phillips, J., and Nicol 2023): - Likelihood profile analysis - Tagger effects evaluation - Mixing period validation
Residual analysis (Day et al. 2023): - Size composition fits - CPUE index fits - Tag recovery patterns - Identify systematic patterns
Extensive testing (Day et al. 2023): - Self-consistency checks for a diagnostic model (possibly including a few key grid models depending on model run time) - Jittering to ensure global optimum
1. Spatial SC19-#4 - 9R vs 6R? - Both in grid?
2. Tag mixing SC19-#14 - Test 1Q, 2Q, 3Q? - All in grid or select one?
3. Growth SC19-#7 - Internal estimation? - Or external priors?
4. Indonesia catch - Accept 2024 spike? - Smooth transition? - Maintain historical?
5. Analysis priorities - CPUE effort creep? SC19-#3 - Yield-per-recruit? SC19-#6
All code and data publicly available:
Docker image for reproducibility:
GitHub Actions for quality control and continuous integration:
Benefits: - Community can review methods - Results fully reproducible - Facilitates review and collaboration - Supports future assessments
All materials publicly available
Analysis: github.com/PacificCommunity/ofp-sam-2026-BET
Report: github.com/PacificCommunity/ofp-sam-2026-BET-report
Pacific Community (SPC)
Oceanic Fisheries Programme
Contact: kyuhank@spc.int
Pre-PAW: 2026 BET Assessment | SPC-OFP